Uploaded March 2026 | Updated November 2026, 3 days ago
From building Electron and helping ship the Slack desktop app to now shaping Claude Cowork at Anthropic, Felix Rieseberg has spent years working at the interface layer. In this episode, Felix joins us to unpack how Claude Cowork emerged from Anthropic’s prototype-first culture, why execution has suddenly become cheap enough that teams can “just build all the candidates,” and why the real frontier in AI products is no longer better chat, but trusted task execution. He also shares why Anthropic is betting on local-first agent workflows, why skills may matter more than most people realize, and how the hardest questions ahead are about autonomy, safety, portability, and the changing shape of knowledge work itself.
We discuss:
• Felix’s path: Slack desktop app, Electron, Windows 95 in JavaScript, and now building Claude Cowork at Anthropic
• What Claude Cowork actually is: a more user-friendly, VM-based version of Claude Code designed to bring agentic workflows to non-terminal-native users
• Anthropic’s prototype-first culture: why Cowork was built in 10 days using many pre-existing internal pieces, and how internal prototypes shaped the final product
• The local machine debate: why Felix thinks Silicon Valley is undervaluing the local computer, and why putting Claude “where you work” is often more powerful
• Why Claude gets its own computer: the VM as both a safety boundary and a capability unlock, letting Claude install tools, run scripts, and work more independently without constant approval
• How Cowork differs from Claude Code: coding evals vs. knowledge-work evals, different system-prompt tradeoffs, longer planning horizons, and heavier use of planning and clarification tools
• Skills vs. MCPs: why Felix is increasingly interested in file-based, text-native interfaces that tell the model what to do, rather than forcing everything through rigid tool schemas
• Real use cases already happening today: uploading videos, organizing files, handling taxes, managing calendars, debugging internal crashes, analyzing finances, and automating repetitive browser workflows
• Why AI products should work with your existing stack: Anthropic’s bias toward integrating with Chrome, Office, and existing workflows instead of rebuilding every app from scratch
• Computer use one year later: how much better it has gotten, why vision plus browser context is such a superpower, and why letting Claude see the thing it is working on changes everything
• The future of junior work: Felix’s concerns about entry-level roles, labor-market disruption, and whether AI can compress early-career learning into denser simulated experience
• Why Waterloo grads stand out: internships, shipping experience, and learning how real teams build products versus purely theoretical academic preparation
• The agentic future of the desktop: what it means for Claude to have its own computer, whether AI should act on your machine or a remote one, and how intimacy with personal data changes the product design space
• Anthropic’s Labs mentality: wild internal experiments, half-broken future-looking prototypes, and the broader effort to move users from asking questions to delegating increasingly long and valuable tasks
Substack Article w/Show Notes: https://www.latent.space/p/felix-anthropic
—
Felix Rieseberg
• X: https://x.com/felixrieseberg
• LinkedIn: linkedin.com/in/felixrieseberg
Anthropic
• anthropic.com
Timestamps
00:00 Cheap execution and building all the candidates
00:44 Intro in the new Kernel studio
02:47 What Claude Cowork is
04:18 Why user-friendly can be more powerful
05:33 How Anthropic built Cowork
07:09 Prototype-first product development
08:00 Why local computers still matter
09:20 Skills, primitives, and platform leverage
12:13 Cowork’s architecture: VM + Chrome + system prompt
15:38 Felix’s own bug-fixing Cowork workflows
17:38 Local-first agents
20:16 Evals, planning, and knowledge-work optimization
23:14 What Anthropic means by evals
24:21 Scaffolding, tools, and why skills matter
27:44 Demo: YouTube uploads and self-generated skills
31:03 Calendar automation and cleaning your desktop
34:47 Browser context and why DOM access matters
37:47 Skills portability and plugins
44:36 Which AI categories survive?
46:19 Junior jobs, simulated work, and labor disruption
52:00 Gradual takeoff vs big-bang takeoff
53:42 Finance, taxes, and enterprise verticals
56:24 Vision and the improvement in computer use
57:31 Why Claude writes its own scripts
58:06 Should Claude have its own computer?
1:01:26 Windows 95 in JavaScript
1:03:19 VM tradeoffs and sandbox design
1:07:23 Approval fatigue and safe delegation
1:11:18 The future of Cowork
1:12:27 What comes next for agentic knowledge work
1:15:13 Electron, Chromium, and desktop software lessons
1:22:16 Multiplayer agents and coworker-to-coworker workflows
1:26:05 Anthropic Labs and closing thoughts
From building Electron and helping ship the Slack desktop app to now shaping Claude Cowork at Anthropic, Felix Rieseberg has spent years working at the interface layer. In this episode, Felix joins us to unpack how Claude Cowork emerged from Anthropic’s prototype-first culture, why execution has suddenly become cheap enough that teams can “just build all the candidates,” and why the real frontier in AI products is no longer better chat, but trusted task execution. He also shares why Anthropic is betting on local-first agent workflows, why skills may matter more than most people realize, and how the hardest questions ahead are about autonomy, safety, portability, and the changing shape of knowledge work itself.
We discuss:
• Felix’s path: Slack desktop app, Electron, Windows 95 in JavaScript, and now building Claude Cowork at Anthropic
• What Claude Cowork actually is: a more user-friendly, VM-based version of Claude Code designed to bring agentic workflows to non-terminal-native users
• Anthropic’s prototype-first culture: why Cowork was built in 10 days using many pre-existing internal pieces, and how internal prototypes shaped the final product
• The local machine debate: why Felix thinks Silicon Valley is undervaluing the local computer, and why putting Claude “where you work” is often more powerful
• Why Claude gets its own computer: the VM as both a safety boundary and a capability unlock, letting Claude install tools, run scripts, and work more independently without constant approval
• How Cowork differs from Claude Code: coding evals vs. knowledge-work evals, different system-prompt tradeoffs, longer planning horizons, and heavier use of planning and clarification tools
• Skills vs. MCPs: why Felix is increasingly interested in file-based, text-native interfaces that tell the model what to do, rather than forcing everything through rigid tool schemas
• Real use cases already happening today: uploading videos, organizing files, handling taxes, managing calendars, debugging internal crashes, analyzing finances, and automating repetitive browser workflows
• Why AI products should work with your existing stack: Anthropic’s bias toward integrating with Chrome, Office, and existing workflows instead of rebuilding every app from scratch
• Computer use one year later: how much better it has gotten, why vision plus browser context is such a superpower, and why letting Claude see the thing it is working on changes everything
• The future of junior work: Felix’s concerns about entry-level roles, labor-market disruption, and whether AI can compress early-career learning into denser simulated experience
• Why Waterloo grads stand out: internships, shipping experience, and learning how real teams build products versus purely theoretical academic preparation
• The agentic future of the desktop: what it means for Claude to have its own computer, whether AI should act on your machine or a remote one, and how intimacy with personal data changes the product design space
• Anthropic’s Labs mentality: wild internal experiments, half-broken future-looking prototypes, and the broader effort to move users from asking questions to delegating increasingly long and valuable tasks
Substack Article w/Show Notes: https://www.latent.space/p/felix-anthropic
—
Felix Rieseberg
• X: https://x.com/felixrieseberg
• LinkedIn: linkedin.com/in/felixrieseberg
Anthropic
• anthropic.com
Timestamps
00:00 Cheap execution and building all the candidates
00:44 Intro in the new Kernel studio
02:47 What Claude Cowork is
04:18 Why user-friendly can be more powerful
05:33 How Anthropic built Cowork
07:09 Prototype-first product development
08:00 Why local computers still matter
09:20 Skills, primitives, and platform leverage
12:13 Cowork’s architecture: VM + Chrome + system prompt
15:38 Felix’s own bug-fixing Cowork workflows
17:38 Local-first agents
20:16 Evals, planning, and knowledge-work optimization
23:14 What Anthropic means by evals
24:21 Scaffolding, tools, and why skills matter
27:44 Demo: YouTube uploads and self-generated skills
31:03 Calendar automation and cleaning your desktop
34:47 Browser context and why DOM access matters
37:47 Skills portability and plugins
44:36 Which AI categories survive?
46:19 Junior jobs, simulated work, and labor disruption
52:00 Gradual takeoff vs big-bang takeoff
53:42 Finance, taxes, and enterprise verticals
56:24 Vision and the improvement in computer use
57:31 Why Claude writes its own scripts
58:06 Should Claude have its own computer?
1:01:26 Windows 95 in JavaScript
1:03:19 VM tradeoffs and sandbox design
1:07:23 Approval fatigue and safe delegation
1:11:18 The future of Cowork
1:12:27 What comes next for agentic knowledge work
1:15:13 Electron, Chromium, and desktop software lessons
1:22:16 Multiplayer agents and coworker-to-coworker workflows
1:26:05 Anthropic Labs and closing thoughts




![Scaling Past Informal AI - Carina Hong, Axiom Math
Carina Hong, founder and CEO of Axiom Math, joins the AI for Science podcast right after closing a $200M Series A to argue that the road to superintelligence runs through formal verification — not as a bug fix, but as the only way to compound and scale AI brilliance. Her company, seven months old and 30 people strong, scored a perfect 120/120 on the 2024 Putnam exam, beating the best human and every other AI system at the time. We dig into the Lean theorem prover, why verified generation gives better training signal than informal RL, the hard specification problem, and why Carina believes an informal system alone can never reach math AGI.
00:00 — [INTRO — spliced from final take at 01:47:28]
00:52 — The $200M Series A and the Math Startup Thesis
04:52 — Verified AI: Scaling Brilliance, Not Fixing Lousiness
13:42 — Axioms System: Lean Data, RL, and the Putnam Perfect Score
22:12 — Mathematical Discovery — Before the Conjecture
25:12 — Rices Theorem, Incompleteness, and Practical Limits
30:42 — Code With Proof — The Verina Benchmark
37:57 — Proof Trees, Context Windows, and Scaling Limits
43:57 — Markets, Moat, and the Business Case ($1.6B valuation)
55:27 — Personal Origin Story: Oxford, UCL Gatsby, Stanford Law
01:00:57 — The Erdos Controversy and the Difficulty of Search
01:06:02 — AlphaZero for Math, Self-Improvement
01:08:47 — Startup Advantage and the OpenAI GPTF Thread
01:13:17 — Axle API — Open Infrastructure for Lean at Scale
01:20:47 — Collaboration, Polymath, and Human Attention as the Bottleneck
01:22:21 — Founding Story — Obsession, Law School, and Julie Zhuo
01:26:17 — The Bigger Vision — AGI, Science, and Transfer Learning
01:35:02 — Bottlenecks, Fragmentation, and the Fields Future Scaling Past Informal AI - Carina Hong, Axiom Math](https://i.ytimg.com/vi/abYcV5LHMG4/mqdefault.jpg)



![[State of Post-Training] From GPT-4.1 to 5.1: RLVR, Agent & Token Efficiency — Josh McGrath, OpenAI
From pre-training data curation to shipping *GPT-4o,* *o1,* *o3,* and now *GPT-5 thinking* and the *shopping model,* *Josh McGrath* has lived through the full arc of OpenAIs post-training evolution—from the PPO vs DPO debates of 2023 to todays RLVR era, where the real innovation isnt optimization methods but *data quality, signal trust, and token efficiency.* We sat down with Josh at *NeurIPS 2025* to dig into the state of post-training heading into 2026: why RLHF and RLVR are both just policy gradient methods (the difference is the input data, not the math), how *GRPO* from DeepSeek Math was underappreciated as a shift toward more trustworthy reward signals (math answers you can verify vs. human preference you cant), why *token efficiency* matters more than wall-clock time (GPT-5 to 5.1 bumped evals _and_ slashed tokens), how *Codex* has changed his workflow so much he feels trapped by 40-minute design sessions followed by 15-minute agent sprints, the infrastructure chaos of scaling RL (way more moving parts than pre-training), why *long context* will keep climbing but agents + graph walks might matter more than 10M-token windows, the *shopping model* as a test bed for interruptability and chain-of-thought transparency, why *personality toggles* (Anton vs Clippy) are a real differentiator users care about, and his thesis that the education system isnt producing enough people who can do *both distributed systems and ML research*—the exact skill set required to push the frontier when the bottleneck moves every few weeks.
We discuss:
* Joshs path: *pre-training data curation → post-training researcher at OpenAI,* shipping GPT-4o, o1, o3, GPT-5 thinking, and the shopping model
* Why he switched from pre-training to post-training: Do I want to make 3% compute efficiency wins, or change behavior by 40%?
* The *RL infrastructure challenge:* way more moving parts than pre-training (tasks, grading setups, external partners), and why babysitting runs at 12:30am means jumping into unfamiliar code constantly
* How *Codex* has changed his workflow: 40-minute design sessions compressed into 15-minute agent sprints, and the strange trapped feeling of waiting for the agent to finish
* The *RLHF vs RLVR debate:* both are policy gradient methods, the real difference is *data quality and signal trust* (human preference vs. verifiable correctness)
* Why *GRPO* (from DeepSeek Math) was underappreciated: not just an optimization trick, but a shift toward reward signals you can actually trust (math answers over human vibes)
* The *token efficiency revolution:* GPT-5 to 5.1 bumped evals _and_ slashed tokens, and why thinking in tokens (not wall-clock time) unlocks better tool-calling and agent workflows
* *Personality toggles:* Anton (tool, no warmth) vs Clippy (friendly, helpful), and why Josh uses custom instructions to make his model just a tool
* The *router problem:* having a router at the top (GPT-5 thinking vs non-thinking) _and_ an implicit router (thinking effort slider) creates weird bumps, and why the abstractions will eventually merge
* *Long context:* climbing Graph Blocks evals, the dream of 10M+ token windows, and why agents + graph walks might matter more than raw context length
* Why the education system isnt producing enough people who can do *both distributed systems and ML research,* and why thats the bottleneck for frontier labs
* The 2026 vision: *neither pre-training nor post-training is dead,* were in the fog of war, and the bottleneck will keep moving (so emotional stability helps)
—
Josh McGrath
* OpenAI: https://openai.com
* https://x.com/j_mcgraph
00:00:00 Introduction: Josh McGrath on Post-Training at OpenAI
00:04:37 The Shopping Model: Black Friday Launch and Interruptability
00:07:11 Model Personality and the Anton vs Clippy Divide
00:08:26 Beyond PPO vs DPO: The Data Quality Spectrum in RL
00:01:40 Infrastructure Challenges: Why Post-Training RL is Harder Than Pre-Training
00:13:12 Token Efficiency: The 2D Plot That Matters Most
00:03:45 Codex Max and the Flow Problem: 40 Minutes of Planning, 15 Minutes of Waiting
00:17:29 Long Context and Graph Blocks: Climbing Toward Perfect Context
00:21:23 The ML-Systems Hybrid: Whats Hard to Hire For
00:24:50 Pre-Training Isnt Dead: Living Through Technological Revolution [State of Post-Training] From GPT-4.1 to 5.1: RLVR, Agent & Token Efficiency — Josh McGrath, OpenAI](https://i.ytimg.com/vi/botHQ7u6-Jk/mqdefault.jpg)

